Guides › Plans for company data

Which ChatGPT or Claude Plan Is Safe for Company Data? Plus, Business, Team and Enterprise Compared

A company needs a business plan at minimum. ChatGPT Free, Go, Plus and Pro and Claude Free, Pro and Max are personal plans under consumer terms. On those plans each user controls whether conversations are used for training, and no administrator can see or change the setting. ChatGPT Business and Claude Team run under commercial terms that exclude your content from training by default. ChatGPT Enterprise and Claude Enterprise add what a security review asks for, which is automated provisioning, audit logs, a compliance API and a retention period you set. Paying for Plus, Pro or Max buys more usage and leaves the terms unchanged.

The answer in four points
  • Personal plans. Free, Go, Plus, Pro and Max sit under consumer terms. ChatGPT uses conversations for training unless the user opts out. Claude uses them only if the user allows it.
  • Business and Team. No training by default, a shared workspace and central billing. Both include SSO. Neither plan includes SCIM, a compliance API or admin-set retention.
  • Enterprise. Adds SCIM, audit logs, a compliance API and a retention period administrators set, on both products.
  • Regulated data. OpenAI offers no business associate agreement for ChatGPT Business. Protected health information needs an Enterprise plan with the HIPAA configuration in place.

Plan by Plan

Each row paraphrases OpenAI's and Anthropic's own documentation as of November 2026. Prices and usage limits are left out because they change often and do not affect the data terms.

PlanTermsTraining on your chatsRetentionAdmin controlsFor company data
ChatGPT Free, Go, Plus, ProConsumerYes, unless the user turns off the model improvement setting.Kept until the user deletes them. Deleted chats are removed within 30 days.None. The account belongs to the individual.No.
ChatGPT BusinessCommercialNo, by default.Deleted conversations are removed within 30 days. OpenAI lists admin-controlled retention as an Enterprise commitment.Workspace admin console and SAML SSO. No SCIM, no Compliance API, no BAA.Yes for internal and confidential business data.
ChatGPT EnterpriseCommercialNo, by default.Set by workspace administrators.SSO, SCIM, role-based access, the Compliance API and logs platform, IP allowlisting, and data residency for eligible customers. A BAA is available with a regulated workspace.Yes for internal and confidential business data. Regulated data needs the matching configuration.
Claude Free, Pro, MaxConsumerOnly if the user allows it in the model improvement setting.Kept until the user deletes them. De-identified data may be kept for up to five years when model improvement is allowed.None. The account belongs to the individual.No.
Claude TeamCommercialNo, by default. Conversations submitted with the feedback buttons can be used.Kept in the product. Custom retention is an Enterprise feature.Admin console, central billing and SSO. No domain capture, SCIM, audit logs or Compliance API.Yes for internal and confidential business data.
Claude EnterpriseCommercialNo, by default. Conversations submitted with the feedback buttons can be used unless Owners turn feedback off.Owners set a custom period of 30 days or more. The default is indefinite.SSO, domain capture, SCIM, audit logs and the Compliance API. A HIPAA-ready configuration with a BAA is available.Yes for internal and confidential business data. Regulated data needs the matching configuration.

A Paid Personal Plan Is Still a Personal Plan

The most common mistake is treating a paid subscription as a business account. ChatGPT Plus and Pro and Claude Pro and Max are sold to individuals under consumer terms, and an expense report does not change that. The account belongs to the employee. The company cannot see what was entered, cannot set retention, cannot confirm the training setting and cannot close the account when the person leaves. Opt-outs also apply going forward only. Anthropic states that data already included in a training run or a trained model stays there. A company that finds personal-account use after the fact cannot fully undo it, which is the argument for providing a sanctioned plan first.

What Business and Team Leave Out

ChatGPT Business and Claude Team solve the contract problem. Your content is excluded from training by default, the provider acts as a processor under a data processing addendum, and the workspace belongs to the company. For a small firm working with internal documents, client correspondence and code, that is a reasonable place to stop. Four things are missing. Provisioning is manual. Both plans support SSO, which blocks sign-in once the identity account is disabled, but no SCIM removes the seat automatically. There is no compliance API to feed a SIEM or an eDiscovery tool. Retention is not set by an administrator. And OpenAI does not sign a business associate agreement for ChatGPT Business, which rules it out for protected health information.

When Enterprise Is Required

  1. You handle regulated data.Protected health information needs a BAA, which sits at the Enterprise tier on both products. See the HIPAA guide.
  2. Security or legal needs the logs.Audit logs and a compliance API are Enterprise features.
  3. Retention has to follow a policy.Only Enterprise lets an administrator set the period.
  4. Offboarding has to be automatic.SCIM removes access when the identity provider does.
  5. Data has to stay in a region.OpenAI offers data residency to eligible Enterprise customers. Ask Anthropic's account team about residency for Claude Enterprise.
  6. Customers ask for evidence.Security questionnaires ask who had access and what was retained. Enterprise is the tier that can answer.

What No Plan Changes

A plan governs what the provider does with content after it arrives. It does not change what arrives. A customer's name and account number sent as context under an Enterprise agreement are still held by a third party, and a connector retrieves whatever the user's identity can open in the source system. Whether that is acceptable depends on the data and your obligations. The guides on putting company data in AI and connector security cover that side.

Where Hardshell Sits

Hardshell addresses what is sent, on any plan. It connects enterprise data to ChatGPT, Claude and other assistants without migrating it, enforces each user's existing permissions at query time, substitutes stand-ins for real identifiers before content reaches the provider, and records every retrieval. The plan decides what the provider may do with the content, and Hardshell decides which identifiers are in it. See how Hardshell works.

Frequently Asked Questions

Is ChatGPT Plus safe for company data?

Not for confidential data. Plus is a personal plan under consumer terms. Conversations are used for training unless the user opts out, and the company has no administrative control over the account.

Does ChatGPT Business train on your data?

No. OpenAI states that content from ChatGPT Business workspaces is excluded from training by default, as it is on Enterprise.

What is the difference between ChatGPT Business and Enterprise for security?

Both exclude your content from training by default. Enterprise adds SCIM provisioning, role-based access, the Compliance API, administrator-set retention and data residency for eligible customers, and it is the tier where a BAA is available.

Is Claude Pro safe for company data?

Claude Pro is a personal plan under consumer terms. Conversations are used for training only if the user allows it, and the company cannot see or enforce that setting.

Does Claude Team train on your data?

Not by default. Anthropic does not train on inputs or outputs from Team, Enterprise or the API. A conversation submitted with the thumbs up or down buttons is the exception, and Owners can turn feedback off for the organization.

Is turning off training on a personal account enough?

No. The setting is controlled by the individual and applies going forward. Retention, access and offboarding stay outside the company's control.